Source-linked AI summary
Travel Mode- and Purpose-Specific Origin-Destination Matrices for England and Wales from Fused Travel Survey and Mobile Network Data
Bowen Zhang, Chen Zhong, Mingfei Ma, James Golding-Graham
TL;DR
Small-area OD evidence segmented by mode and purpose remains limited, despite its relevance to transport planning. The paper fuses mobile-network movement patterns with survey and population evidence to construct national MSOA matrices. Calibration brings mode-share and purpose controls close to their NTS targets, but the checks do not establish accuracy against independent small-area measurements.
Problem
Transport planning needs up-to-date small-area OD evidence by mode and purpose, but existing sources do not provide those dimensions together.
Source: §1 · pp. 1–2Method
The pipeline preserves mobile-network spatial movement patterns while calibrating all-age, mode and purpose composition against population and survey evidence.
Source: §1 · p. 2, §3.1 · p. 5, Step 1: Standardising geography and… · p. 5Results
The mean absolute mode-share gap across 427 region–band–mode cells falls from 5.07 to 0.01 percentage points, and purpose control groups match their NTS0502 shares within 0.01 percentage points after raking.
Source: §4 · p. 9Takeaways & Limitations
Analysts can extract mode-, period- and, for weekday morning peaks, purpose-specific flows for questions such as school access, corridor work and public-transport appraisal.
Source: §5 · p. 11No independent small-area OD measurement is available for England and Wales, so the reported validation consists of consistency checks rather than a verdict on accuracy.
Source: §4 · p. 9
Abstract
from arXiv · showhide
Origin-destination (OD) matrices sit behind much of quantitative transport planning, from model calibration and accessibility analysis to the appraisal of new services and development. The increasing emphasis on place-based solutions requires mobility data that can support decision-making not only at the strategic level, but also at finer spatial scales. This requires up-to-date OD evidence at small-area resolution, disaggregated by travel mode and purpose, which remains either inaccessible or unavailable. In this work, we present dense MSOA-to-MSOA OD matrices for England and Wales, segmented by seven travel modes and representative time periods, with eight trip purposes for the weekday morning peak. The matrices are built by calibrating aggregate mobile network data provided by BT against National Travel Survey (NTS), census and trip-rate evidence, preserving the observed spatial structure of movement while referencing its age, mode and purpose composition to the survey. The open-source data processing pipeline is released alongside the matrices, so that the construction of the dataset can be inspected in full and adapted to other years, regions or assumptions.
1 Introduction
Local transport planning needs small-area OD evidence segmented by mode and purpose, but existing survey, census and mobile sources each lack dimensions needed for that task. The paper fuses these sources into national MSOA-to-MSOA matrices with mode, time and weekday morning-peak purpose detail, alongside a reproducible pipeline.
1 Introduction
- Local transport decisions need small-area OD evidence because travel-to-work areas often cross administrative boundaries, and policy calls for place-based movement assessments [Department for Transport, 2026].Source: §1 · pp. 1–2
- Analysts rarely obtain mode and purpose detail together and instead choose among sources with different strengths.Source: §1 · p. 2
- Census journey-to-work flows are spatially detailed but cover one purpose and are collected once a decade, while the household NTS lacks the sample size for small-area OD estimation [Office for National Statistics, 2023] [Department for Transport, 2024a].Passive mobile data observe population-scale movement but do not record purpose, infer rather than report mode, and cover device-carrying adults.Source: §1 · p. 2
- The fused release provides dense England-and-Wales MSOA-to-MSOA matrices for seven modes and representative periods, with eight purposes for the weekday morning peak.The published pipeline supports inspection, reproduction and reruns for other years or study areas.Source: §1 · p. 2
2 Data and code release
The release provides MSOA-level OD matrices for England and Wales across two time groupings and seven modes, with purpose-specific matrices only for weekday AM peaks. It also includes pipeline documentation and metadata, while purpose allocation and the source data impose specific scope conditions.
2 Data and code release
- The matrices are CSV files at MSOA level for England and Wales, with cell values reporting trip volumes across seven modes.Table 1 identifies released products and their processing stages; Table 2 gives the equivalent long-form schema, while dimensions are encoded in file and folder names.Source: §2 · pp. 2–3, Table 1 · p. 3, Table 2 · p. 4
- The representative-week grouping is an average seven-day total, whereas weekday AM peak reports an average single weekday 07:00–09:59 period.Periods are assigned by departure time from the origin; mode targets use the 2024 NTS and purpose targets the 2023–24 NTS, both covering England only.Source: §2 · p. 3
- The release contains 70 matrices: seven representative-week all-purpose files, seven weekday AM-peak all-purpose files, and 56 weekday AM-peak mode-and-purpose files.Summing the eight purpose matrices for a mode recovers its all-purpose AM-peak matrix.Source: §2 · p. 3
- Purpose-specific matrices are limited to the weekday AM peak because trip-production rates count activities, not directional trips, and do not record when return journeys occur.Most morning-peak trips are outbound, which keeps but does not remove this mismatch.Source: §2 · p. 3
- Each calibration step produces diagnostic files, which are not distributed but are regenerated when users rerun the pipeline.Source: §2 · p. 3
3 Methods: Dataset construction and fusion
Five input streams feed a configurable, four-step pipeline that calibrates BT mobile aggregates against population, survey, and trip-rate evidence. It adds age, mode, time, and purpose dimensions while retaining the observed OD structure.
3 Methods: Dataset construction and fusion
- The four-step pipeline is configurable, and only its BT mobile input is commercially licensed; the other four inputs are open or available under standard licences.Rebuilding outside the original project requires a licence for a compatible mobile feed.Source: §3 · p. 4, §3.1 · p. 5, §3.2 · p. 9
Step 1: Standardising geography and building an all-age base
- Adult BT volumes are uplifted to all ages using origin populations, with added child travel inheriting the adult OD pattern at that origin.The national mean uplift is 1.226, corresponding to a mean origin child share of 18.3%.Source: Step 1: Standardising geography and… · p. 5
Step 2: Mode-share calibration
- Within region–trip-length groups, mode calibration adjusts record volumes to survey-referenced mode shares without moving trips between OD pairs.Groups without survey targets retain their mobile-base mode composition; in the national build these are 25 Welsh groups.Source: Step 2: Mode-share calibration · p. 6
Step 3: Road-mode split and mode–time refinement
- The road-mode split uses NTS shares to divide ROAD into cycle, private car, motorcycle, and bus, while mode–time refinement renormalizes observed groups and conserves origin-period totals.Cells without NTS road evidence default to private car; no travel is assigned to a mode absent from an origin-period cell.Source: Step 3: Road-mode split and mode–time… · p. 6
Step 4: Purpose allocation and raking to survey controls
- Purpose shares come from origin-specific population segments weighted by trip-production rates, then iterative raking aligns weekday allocations with NTS0502 controls while preserving row totals.Weekend periods retain local-prior shares, and purpose allocation does not move trips between OD pairs.Source: Step 4: Purpose allocation and raking… · pp. 7–8
3.2 Implementation
- The implementation is an open-source Python package with tested, configurable behavioral steps, allowing builds for other regions, years, or assumptions to differ by configuration.The national build enabled child uplift, the road-mode split, mode–time constraint, and NTS0502 calibration.Source: §3.2 · pp. 8–9
4 Validation
The checks show that mode and purpose calibration closely matches its survey controls, but they are consistency checks rather than independent evidence of accuracy.
4 Validation
- Because calibration uses the same survey targets assessed by these checks, they show that adjustments behave as specified, not that matrices match travel on the ground.Purpose is inferred rather than observed, and the authors identify no independent small-area OD measurement for England and Wales.Source: §4 · p. 9
- Mode-share mean absolute gaps fall from 5.07 to 0.01 percentage points, or from 7.86 to 0.01 when weighted by cell volume.These comparisons cover the 427 region–band–mode cells with NTS targets; Figure 1a shows the mode-share comparison.Source: §4 · p. 9, Figure 1 · p. 10
- Private-car gaps fall from 13.5 to 0.03 percentage points and walking gaps from 11.6 to below 0.01; aggregate road and walking shares move to 65.4% and 31.0%.One cell remains above a 0.5-point gap because the survey reports a mode absent from the mobile data; the weekday AM-peak product is later refined to trip-rate mode–time splits.Source: §4 · p. 9
- Raking brings every weekday AM-peak purpose control within 0.01 percentage points of NTS0502, while origin-level purpose composition retains the local prior.Before raking, commuting is 35.7% against 20.4%; personal business is 7.3% against 15.6%, and holiday travel is 2.4% against 10.2%.Source: §4 · pp. 9–10
5 Discussion and conclusion
The matrices let analysts extract flows by mode, period and, for weekday morning peaks, purpose. Their interpretation remains bounded by source and inference limitations, while the configurable pipeline supports extensions and rebuilding.
5 Discussion and conclusion
- Mode- and period-specific matrices, with weekday AM-peak purpose splits, let analysts extract flows for school access, corridor work or public-transport appraisal.Examples include education trips, car flows, and bus and rail volumes.Source: §5 · p. 11
- The matrices are synthetic planning estimates, and calibration cannot correct biases that do not appear in its region-and-trip-length margins.Welsh mode shares lack NTS calibration; child volumes inherit adult OD patterns, and purpose is inferred rather than observed.Source: §5 · p. 11
- The pipeline can be rebuilt for another year, region or assumptions; destination-side employment or retail evidence could make purpose respond to destinations as well as origin populations.The preserve-structure-then-calibrate workflow is described as applicable to licensed OD sources with published behavioural margins.Source: §5 · p. 11
Data availability statement
The matrices and release metadata are deposited, and the processing pipeline is available under an open-source licence. Rebuilding and finer-resolution releases have specific data and disclosure constraints.
Data availability statement
- The published matrices and release metadata are deposited, and the uk-travel-pipeline source code is available under the MIT licence.The data are available through Zenodo and the code through GitHub.Source: Data availability statement · p. 12
- Rebuilding requires an equivalent licence for a compatible BT mobile network feed, which is commercially licensed and not redistributed.Four of the five inputs are open or available to researchers under standard licences.Source: Data availability statement · p. 12
- The aggregate synthetic product is released at MSOA level; finer geography or time resolution would require a new disclosure assessment.Source: Data availability statement · p. 12
Funding
The research reports support from the European Research Council and the Economic and Social Research Council, UK.
Funding
- Support came from the European Research Council under Horizon 2020 and from the Economic and Social Research Council, UK.ERC grants No. 949670 and No. 101212775; ESRC grant No. ES/Y010558/1.Source: Funding · p. 12